Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

LaMDA: Google's New Conversation Technology

blog.google

Post reply on HN
Print view View original

141–150 of 157 posts

Re: LaMDA: Google's New Conversation Technology

#141

Post by dougmwne »

Mizza wrote: post #27
moralestapia wrote: post #24

>“I just started taking guitar lessons.” >You might expect another person to respond with something like: >“How exciting! My mom has a vintage Martin that she loves to play.” Not really IMO, I think they just pushed it to just say "whatever's related to the topic". People who behave like that irl come off as self-absorbed and obnoxious most of the time, hope future AIs are not like that!

I came here to point this out - this is how people from California communicate. They simply take turns talking about themselves, orbiting around the same subject until somebody says something with enough gravity to drag the conversation into its next orbit. It was really off-putting when I first arrived, I got used to it over time, though I still don't like it and it's sad to see it used as the default model for this…

Oh my. This so much. To be honest, it came across as incredibly selfish and self-centered when I was there, as a person from the east coast. It was as if people talk at each other instead of with each other.

One element of this I think is that California is highly individualistic. The people there ARE self-absorbed. It is a celebrated cultural value. It's the exact mirror of Tall Poppy Syndrome. No one cuts you down for talking about your own interests and uniqueness instead of collectivist and egalitarian topics like sports or the weather.

There's also a less intricate social code that comes from mixing together many people from different cultures. In a high-context, temporarily stable society, things get stratified and the litte social lubricants, expectations, and niceties get written into the socal contract and within the structure of the language itself (formal vs. informal forms). In a low-context and dynamic society, you are basically free to communicate without referencing the rulebook on what you should be saying or how you should be saying it and you bring less expectation that the other party will respond in some particularly scripted way.

Re: LaMDA: Google's New Conversation Technology

#142

Post by royal_ts »

There's a clear decline in understanding what's being said with Google assistant and it's like this for a while now. I'm using this stuff since 2010ish when it was still called Google Now and it could do more for you back then.

Nowadays it doesn't know what I mean when I ask "what bus do I have to take to get to Dave". First of all, it doesn't know what to do with contacts anymore. Second: you have to say "what public transportation do I have to take to get to [Dave's location]". It just doesn't know that a bus or tram is interchangeable with public transport. Another example is asking for the last train home. Nowadays it will list you like 3 trains trains from now on but not the last one which would be hours away. Asking for when the next train comes is the same, it knows the answer but shows a list instead without saying anything.

It just gets worse and worse so that the only thing which works reliable is asking for the weather and setting alarms - which works like this for over 10 years.

I have a very hard time believing that this will result in a product we can use and if so: it will be in English only forever. Speaking of english only: their live translation overlay and their recorder app were some selling points for the Pixel at this time and to this day it's only available in English and that's it. Same for their "better artificially generated voices"

It's only marketing and I don't believe we will get anything out of this ever and if we do it's nothing like this at all.

Re: LaMDA: Google's New Conversation Technology

#143

Post by dougmwne »

onlyrealcuzzo wrote: post #64
moralestapia wrote: post #40

Earlier quoted context omitted.

"Hey, that's nice", "How's that going?" or something like it. Also, no response is also a response, you could just listen to them. No one's going to say "I just started taking guitar lessons" and remain silent for the rest of the evening; for sure they have much more to say about it.

Wait. Why is it self-absorbed for someone to include themselves in the conversation, but it's not self-absorbed for you to come up to me and talk to me about your guitar playing? You can come up to me and say, "Hey, guess what!? I just started playing guitar!" And I can't say, "No kidding? My mom just started playing guitar, too!" or, "No way! I just started drumming a few weeks ago. Maybe we can jam soon?" How is my…

I would say that in the best case we would take turns sharing about something personally important and being listened to and engaged with where we're at instead of quickly pulled off in another direction. For a bit anyway, before you get the ball.

I think the core of this question is generosity of time and attention, each person's ability to suspend their solipsism and engage with the other person as a person of equal importance to themselves. A good conversation is an act of intimacy.

We all want to be heard but no one ever seems to want to do the listening.

Re: LaMDA: Google's New Conversation Technology

#144

Post by bserge »

andrewtbham wrote: post #15

I have been researching making a therapy chat bot for years... it's going to happen. https://github.com/andrewt3000/carl_voice

I made my own, but it only works for me heh. I know myself really well, I know what I need, so it's a somewhat simple task. Creating an actually useful one that anyone can install and use seems much, much harder.

I like Google's TTS and it was incredibly annoying to integrate with my bot. I realize they don't want you using it for free for your own purposes, but damn.

Re: LaMDA: Google's New Conversation Technology

#145

Post by bserge »

wing-_-nuts wrote: post #16
andrewtbham wrote: post #15

I have been researching making a therapy chat bot for years... it's going to happen. https://github.com/andrewt3000/carl_voice

To be honest, the thought of AI doing therapy (and it's reliance on big data) sounds very black mirror to me. How long do you think it'll take ad companies to do an end run around patient protections and start serving ads based on "metadata"? Got anxiety? Try our new meds! Worried about the future? How about some life insurance or annuities? Etc. Things will get very exploitative very fast.

It would have to be offline and highly personalized by a professional (or by the individual), I see no other way around it.

Re: LaMDA: Google's New Conversation Technology

#146

Post by bserge »

ehsankia wrote: post #106
practicalpants wrote: post #105

Solving conversational intelligence is the same as solving general intelligence (AGI), so unless google has solved the latter I would expect any chatbot system to still be pretty annoying for any nuanced or real world conversation.

Not necessarily. GPT-3 for example can put out entire articles, even generate images or generate valid HTML, doesn't mean it has reached AGI, and while it's far from perfect, it can still have utility. As per the demo, I can imagine learning about a topic through conversations with an AI. It would still be regurgitating existing knowledge, but in a conversational back and forth way, which to some can have value.

It would be very nice if it was hooked into Google Search in real time. I guess that's the next step for an actually useful Google Assistant?

(GPT?) auto generated replies for karma building seem to be gaining traction, I wonder if it can spell the death of forums haha (most likely, it will bring in better moderation systems).

Re: LaMDA: Google's New Conversation Technology

#147

Post by andrewtbham »

bserge wrote: post #144
andrewtbham wrote: post #15

I have been researching making a therapy chat bot for years... it's going to happen. https://github.com/andrewt3000/carl_voice

I made my own, but it only works for me heh. I know myself really well, I know what I need, so it's a somewhat simple task. Creating an actually useful one that anyone can install and use seems much, much harder. I like Google's TTS and it was incredibly annoying to integrate with my bot. I realize they don't want you using it for free for your own purposes, but damn.

how did you make your own? gpt3?

Re: LaMDA: Google's New Conversation Technology

#148

Post by bserge »

andrewtbham wrote: post #147
bserge wrote: post #144

Earlier quoted context omitted.

I made my own, but it only works for me heh. I know myself really well, I know what I need, so it's a somewhat simple task. Creating an actually useful one that anyone can install and use seems much, much harder. I like Google's TTS and it was incredibly annoying to integrate with my bot. I realize they don't want you using it for free for your own purposes, but damn.

how did you make your own? gpt3?

No, nothing complicated, just pre-made, randomized feedback (so it doesn't sound exactly the same every time) for when I feel real bad. Triggered by multiple keywords when I speak.

It's like talking to myself, which is what I've always done and it's always the same conversations. So I automated it.

Now half+ of it comes from outside, which seems to work better even if it's a synthetic voice with messages I made.

I also used it for nagging medication reminders, as I kept forgetting what, how and when to take (at one point it was... quite a lot, on an overly complicated schedule).

Re: LaMDA: Google's New Conversation Technology

#149

Post by zitterbewegung »

missingsteps wrote: post #137
zitterbewegung wrote: post #17

If you want a freely available system that attacks this problem then Blenderbot is a good choice (made by Facebook)[1] . It is also freely available and integrated with Huggingface. [1] https://ai.facebook.com/blog/state-of-the-art-open-source-ch... [2] https://huggingface.co/facebook/blenderbot-3B

It would be interesting to hear how the google and facebook bots discuss who is the winner in chatbot's technology, and I wonder if the evaluator function of both model agree on who is the overall winner, this imply some scale or measure to divide the knowledge graph into brachs and compare the quality of conversational technology of both models, perhaps using a random forest model to evaluate the performance mean sc…

Wolfram has no chatbot and is a data source for chatbots. Blenderbot has no memory across conversations. I’m not familiar with transfer learning or hyper parameter search for transformers when primarily finetuning is used .

Re: LaMDA: Google's New Conversation Technology

#150

Post by dlkf »

I'm surprised at how much money is being spent micro-optimizing the superficial and contingent aspects of intelligent behaviour. It's as though you've got a toddler who has no idea how to play piano, and instead of trying to teach them a C-scale, you're investing in increasingly sophisticated studio equipment to record random cacophony.
Post reply on HN